What is DeepForest?

DeepForest is a python package for training and predicting individual tree crowns from airborne RGB imagery. DeepForest comes with a prebuilt model trained on data from the National Ecological Observatory Network. Users can extend this model by annotating and training custom models.

_images/image.png

What’s new in DeepForest 1.0.0?

The original project was written in tensorflow based on keras-retinanet. When tensorflow updated to 2.0 there were breaking changes and the authors of keras-retinanet decided to not attempt to recover the project. The rapid development of open machine learning community resources means that tensorflow 1.14.0, which is required for deepforest, will rapidly become out of date. Fear not, starting in 1.0.0, the model now depends on the pytorch and torchvision and will be stable for the forseeable future.

How does deepforest work?

DeepForest uses deep learning object detection networks to predict bounding boxes corresponding to individual trees in RGB imagery. DeepForest is built on the retinanet model from the torchvision package and designed to make training models for tree detection simpler.

For more about the motivation behind DeepForest, see a recent talk I gave at the Florida Musuem on Natural History

Demo

For a quick example of model performance on sample images, see our demo shiny app. Please note that the model continues to improve and the app model may not reflect results from the current release.

http://weecology.westus.cloudapp.azure.com/neon/

License

Free software: MIT license

Why DeepForest?

Remote sensing can transform the speed, scale, and cost of biodiversity and forestry surveys. Data acquisition currently outpaces the ability to identify individual organisms in high resolution imagery. Individual crown delineation has been a long-standing challenge in remote sensing and available algorithms produce mixed results. DeepForest is the first open source implementation of a deep learning model for crown detection. Deep learning has made enormous strides in a range of computer vision tasks but requires significant amounts of training data. By including a trained model, we hope to simplify the process of retraining deep learning models for a range of forests, sensors, and spatial resolutions.

Feedback

All issues can be submitted to the github repo. We look forward to hearing about the performance of the prebuilt and custom models.